Face recognition using CNN and siamese network

Abstract

Facial recognition is no longer a cutting-edge technology; it is now a part of everyday life. It has been used for various security and profiling applications around the world. Early face detection models were developed during the 1960s and were used to just classify photos of people. In past decades, the face recognition models were optimized and reengineered to identify all the people in each frame of real-time, high-resolution video input. It still has a wide variety of applications to be implemented and can be further optimized for high precision using different approaches. In this study, we have implemented two different approaches for facial detection. The first is a CNN-based approach that extracts keypoints from an image and classifies it using a KNN algorithm. The next approach uses a Siamese network to classify the input image. The initial part focuses primarily on data collection and training. The following part clearly explains the implementation of both approaches. The performance of these approaches was also evaluated and illustrated optimally.

Keywords

• Facial key-points
• K-nearest neighbor
• Siamese network
• Facial image verification
• Convolutional Neural Network